From owner-freebsd-stable Mon Aug 19 8:33:48 2002 Delivered-To: freebsd-stable@freebsd.org Received: from mx1.FreeBSD.org (mx1.FreeBSD.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id 6E19137B400 for ; Mon, 19 Aug 2002 08:33:45 -0700 (PDT) Received: from mail.dt.e-technik.uni-dortmund.de (krusty.dt.E-Technik.Uni-Dortmund.DE [129.217.163.1]) by mx1.FreeBSD.org (Postfix) with ESMTP id B3C4843E65 for ; Mon, 19 Aug 2002 08:33:44 -0700 (PDT) (envelope-from matthias.andree@gmx.de) Received: from m2a2.myip.org (localhost [127.0.0.1]) by mail.dt.e-technik.uni-dortmund.de (Postfix) with ESMTP id 2905CA3831 for ; Mon, 19 Aug 2002 17:33:42 +0200 (CEST) Received: by merlin.emma.line.org (Postfix, from userid 500) id B5F79507B7; Mon, 19 Aug 2002 17:33:38 +0200 (CEST) To: freebsd-stable@freebsd.org Subject: Re: root (/) not soft-updates by default ? References: <20020819144928.GA6628@nebula.wanadoo.fr> <20020819105452.A14530@blackhelicopters.org> In-Reply-To: <20020819105452.A14530@blackhelicopters.org> (Michael Lucas's message of "Mon, 19 Aug 2002 10:54:52 -0400") From: Matthias Andree Date: Mon, 19 Aug 2002 17:33:38 +0200 Message-ID: Lines: 44 User-Agent: Gnus/5.090008 (Oort Gnus v0.08) Emacs/21.1 (i686-pc-linux-gnu) MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-15 Content-Transfer-Encoding: quoted-printable Sender: owner-freebsd-stable@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.ORG Michael Lucas writes: > Soft updates does not work well with IDE controllers with write > caching enabled; you can cause data loss. > > Most IDE controllers ship with write caching enabled. That's plain wrong, IDE controllers don't bother. If at all, their BIOS would, but why should someone bloat the scarce BIOS space with write cache treatment? It's the ATA drives that ship with the write cache enabled, but that is of no matter to FreeBSD. FreeBSD-4 in its default configuration ENABLES the write cache nonetheless (even if you used e. g. the IBM feature tool to turn it off), unless you put hw.ata.wc=3D"0" into /boot/loader.conf.local or /boot/loader.conf and reboot. (It might be useful to teach atacontrol about the write cache, but then, you don't turn that write cache knob very often.) FreeBSD will still enable the write cache for drives that do tagged queueing if you put hw.ata.tags=3D"1", but with tags, that's a different issue because you can force cache flush so to support fsync(2), for example. > We won't, by default, ship systems which have an unacceptable risk of > data loss. :-) Daring words in the face of current FreeBSD 4 behaviour (that's described above), which runs in "data loss encouraged" mode by default. I have been told S=F8ren Schmidt wanted to change the hw.ata.wc cache setting to default to 0 but was told not to for performance reasons and POLA and things. However, the risk is with the write cache of the drive, not with the softupdates async scheme. The problem is that the write cache can cause reordering writes which must not be reordered. Without write cache, such things don't happen. The write cache can corrupt your file system regardless of whether softupdates are enabled or not. --=20 Matthias Andree To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message